B. COLLECTION OF INFORMATION EMPLOYING STATISTICAL METHODS
1. Describe (including a numerical estimate) the potential respondent universe and any sampling or other respondent selection method to be used. Data on the number of entities (e.g., establishments, State and local government units, households, or persons) in the universe covered by the collection and in the corresponding sample are to be provided for the universe as a whole and for each stratum in the proposed sample. Indicate expected response rates for the collection as a whole. If the collection, was conducted previously, include the actual response rate achieved during the last collection.
The target population is all farms. A farm is defined as any operation from which $1,000 or more of agricultural products were produced and sold or would normally be sold during the reference year (2017). This definition has been used since the 1974 Census of Agriculture (conducted by the Census Bureau). NASS maintains a list sampling frame containing names and addresses of operations qualifying as farms under this definition. The list frame is continuously updated and supports the agricultural estimates program as well as the Census. Processes are in place to identify and eliminate duplication and inactive operators (e.g., deceased, retired, and out-of-business farmers), and to evaluate outside list sources to find new and missing farm operators.
The list building effort leading up to the 2017 Census of Agriculture started the year following the previous Census. During the five year time span between the two censuses NASS performs intensive list building efforts to improve list coverage. List building is completed using two strategies: large general lists of potential farm operators available for all states and state-specific lists of targeted commodities.
The large general lists for all states that NASS processes each year include farm operators who utilize the Farm Service Agency, Risk Management Agency, and those that file federal tax forms related to agricultural production with the Internal Revenue Service. These lists are procured at the national level and produce a high success rate when matched against the NASS list sampling frame.
The other list sources vary by state. States target list sources for specific crop or livestock commodities. The targeted commodities include major commodities as well as commodities which NASS considers to be underrepresented on the list frame. Examples of major commodity list sources include cattle producer lists from the U.S. Cattlemen’s Association, hog manure management plans from the Iowa Department of Natural Resources, and pesticide applicator lists in California. Smaller list sources procured in the past include lists of goat operators, beekeepers, and maple syrup taps. These lists have varying degrees of success in identifying farm operators but are considered a necessity to building a representative list sampling frame.
The National Agricultural Classification Survey (NACS) will be used to screen approximately 1.2 million of these records to determine if they have any agricultural production and should be included on the census mail list. The list development effort is expected to produce a list of approximately 3.2 million names and addresses of farm operators and potential farm operators for the Census.
Smaller or less complex operations that are likely to be able to skip many sections of the full Census questionnaire will receive a shortened version of the questionnaire. If the operations are found to have certain commodities, such as vegetables, fruits, tree nuts, berries, horticulture crops, sheep, hogs, poultry, or aquaculture, they may be re-contacted by phone and asked to complete the part(s) of the questionnaire that were omitted in the shorter version. NASS estimates that 450,000 operators will receive the shorter questionnaire. Based on testing, approximately 15 percent of these operations will be re-contacted.
Response to the Census is required by law under the “Census of Agriculture Act of 1997,” Pub. L. No. 105-113 (7 U.S.C. 2204g). The overall response rate for the 2012 Census of Agriculture was 80.1 percent. Response rates will be monitored by county and low response counties will be targeted for follow-up.
2. Describe the procedures for the collection of information including:
• statistical methodology for stratification and sample selection,
• estimation procedure,
• degree of accuracy needed for the purpose described in the justification,
• unusual problems requiring specialized sampling procedures
The goal of the estimation procedure for the Census is to produce agricultural totals for publication that are fully adjusted for list under-coverage, non-response, and misclassification at the county level. As was done in 2012, a capture-recapture methodology will be used for the adjustments. A paper written by Linda J. Young, Andrea C. Lamas, and Denise A. Abreu entitled The 2012 Census of Agriculture: A Capture-Recapture Analysis, which is currently under review by the Journal of Survey and Statistical Methodology, describes these procedures.
The primary assumptions in capture-recapture are (1) that the population is closed to the formation and deletion of farms between the two sampling periods, (2) that the probability of capture is the same for farms with a given set of characteristics, (3) that the probability of capture does not vary with sample, and (4) that being included in the JAS sample does not affect the probability of responding in the census relative to farms not included in the June Area Sample (JAS) survey, which is approved under OMB # 0535-0213.
With the six-month time between data collection for the JAS and the Census, the population will have had some new farms formed and others that have gone out of business. Kendall (1999 Ecology 80: 2517-2525) showed that, if the farms enter and leave the population at random (violation of (1)), the estimates are still unbiased, but the precision is decreased. Further, if the capture probabilities vary with time (violation of assumption (3)), the estimates continue to be unbiased.
NASS assumes that being included in the JAS sample does not affect the probability of response on the census. With only two samples, it is not possible to test this assumption. If responding to the JAS, increases the probability of response to the census relative to farms that do not respond to the JAS, the estimate of the number of farms will be biased downwards; otherwise, if the probability of response to the census decreases if a farm is in the JAS sample relative to other farms not in the sample, the estimate of the number of farms is biased upwards.
The capture probabilities do vary with type of farm (assumption (2)). To account for this variation in catchability in the models for coverage, response, and misclassification of CML farms, stepwise weighted logistic regression with cross validation will be used for model development, with the selection made from the variables reported on the census. The key reporting variables (State, land in farm (in acres), operator’s sex, operator’s age, operator’s race/ethnicity, type of farm, and total value of production) will be included in all models.
To implement this method these two surveys (JAS and Census) are used. It is assumed that the surveys are independent and that farms with a given set of characteristics are equally likely to be captured on both surveys. NASS uses the June Area Survey (JAS) to estimate the number and types of farms not on the Census Mailing List (CML). The tracts in the JAS that are not on the CML are said to be in the “Not on the Mail List” (NML) domain. If a tract in the NML domain is determined to be a farm during the census, it is an NML farm. The CML is used with the NML in the capture-recapture framework to represent all farming operations across all states in the JAS sample. Although much effort is expended to make the CML as complete as possible, inevitably the CML will not include all U.S. farms, resulting in list under-coverage. Some farm operators who are on the CML will not respond to the Census, despite numerous attempts to contact them. In addition, although each operation is classified as a farm or nonfarm based on their responses on the Census questionnaire, some will be misclassified (some non-farms will be classified as farms and some farms will be classified as non-farms).
Probability estimates for missing a farm due to nonresponse, misclassification, and under-coverage are obtained via logistic regression models using matched records from the CML and JAS. These probability estimates are used in the capture-recapture model to adjust weights. The weights for these estimates are general purpose in that they do not provide any control over expected levels of commodity production. To address this a calibration algorithm is used to create the final set of weights. The calibration algorithm adjusts the weights to align indications with previously published estimates and to reduce bias. (See the attached Calibration for the Census of Agriculture.) The calibration algorithm outputs integer weights to ensure that totals for all sub-domains and cross-tabulations will balance in the final data products.
The NASS area frame, which is used for the independent June Area Survey (JAS), covers all land in the U.S (excluding Alaska) and includes all farms. The land in the U.S. is stratified by characteristics of the land. A probability sample of segments is drawn within each stratum for the JAS. The JAS sample of segments is allocated to strata to provide accurate measures of acres planted to widely grown crops, farm numbers, and inventories of cattle. In June 2017, the operational sample will be increased to improve the farm counts for operations that produce specialty commodities or have socially disadvantaged or minority operators. The supplemental sample will also enhance the capture-recapture modeling.
Prior to the beginning of data collection, input will be solicited from field office staff with the specific objective to identify a small portion of records that need to be specially handled (tagged records). These records may be matches to other surveys, have multiple locations or unique operator structures, have existing data collection plans, or need to have data collected by statisticians or supervisory enumerators, many are typically large or complex operations which we have existing relationships and contact frequently.
In an effort to reduce response burden NASS coordinates data collection activities with concurrent surveys. A majority of these census questionnaires are matched with records in the Agricultural Resource Management Survey III (ARMS III), (OMB approval number 0535-0218) sample. The ARMS III survey collects whole farm data and has many similarities to the census questionnaire. NASS and the Economic Research Service (ERS) have collaborated on an ‘integrated’ questionnaire that incorporates all the necessary Census data items into the ARMS III questionnaire. Effectively, the ARMS III data collection will run concurrent with most of the Census period, with operations in the ARMS III sample targeted for face-to-face enumeration. Upon conclusion of ARMS III data collection, the necessary data captured in ARMS III will be used for the Census of Agriculture.
Some other tagged records will be of farms and ranches that have multiple locations and can be reported by one headquarters operation.
In the population census, which is conducted by the Census Bureau, households are treated equally, by this we mean that there are no adjustments for size for a larger home vs. a smaller home. However, with the Census of Agriculture we are dealing with farm or ranch operations. A small number of farmers could represent a very large portion of production of a certain commodity. These records may be tagged so that NASS staff can be sure that theses questionnaires are completed which may involve personal enumeration.
Finally, there are a very minimal number of records that we have a long history with and have an established enumerator contact. To honor an agreement, NASS staff tags these records for collection by that enumerator.
The consequences of normal handling of these records could manifest itself into increased response burden due to the potential need for follow-up contacts to verify data.
During 2014, 2015, and 2016 NASS staff worked with minority population groups to obtain lists of potential agriculture producers to receive a National Agricultural Classification Survey questionnaire (OMB No. 0535-0140). This effort will continue into 2017. During the collection period if an operation is discovered within one of the under-represented populations, NASS will include their report in the aggregate data.
NASS intends to continue efforts to obtain Census data from individual American Indian farmers and ranchers on reservations in the 2017 census. Historically, the Bureau of the Census and NASS treated most American Indian reservations in the U.S. as single farming operations for the Census. A single Census report was obtained for the entire reservation, including data for any tribally operated farm or ranch and all individual farms and ranches. In 2007, NASS expanded its efforts to reach individual American Indian farms and ranches, on and off reservations. For the majority, individual operators were represented in the Census data in all states. In a few instances operator counts were obtained from reservations which preferred to report aggregated reservation data. Similar procedures will be used in 2017. NASS will also use a customized questionnaire for American Indian operations in the southwestern United States.
3. Describe methods to maximize response rates and to deal with issues of non-response. The accuracy and reliability of information collected must be shown to be adequate for intended uses. For collections based on sampling a special justification must be provided for any collection that will not yield "reliable" data that can be generalized to the universe studied.
In an effort to reach typically under-covered populations, NASS has and will continue to work with community based organizations (CBOs) to recruit and train members of the under-covered populations as enumerators. NASS also plans to send staff to attend events held by CBOs leading up to and throughout the Census data collection period. These staff will provide assistance in filling out Census questionnaires.
Farms that potentially represent a large proportion of the total for a given commodity will be identified prior to data collection and categorized as “must case” records. These will be the top 0.1% to 1.5% depending on the number of farms with the commodity. Non-response from these operations often cannot be accurately adjusted for, so special efforts will be made to encourage response. This may include sending a pre-survey letter to confirm the operation’s address or having a NASS representative visit the operation prior to mailing out Census questionnaires. During the data collection phase, these “must cases” are tracked and will be among the first to receive phone follow-up or personal enumeration.
NASS places a high priority on obtaining comprehensive and uniform coverage of all farms. In order to ensure sufficient coverage in every county, NASS will utilize a Computer Assisted Telephone Interview (CATI) program specifically designed to target records in counties with lower coverage rates. Personal enumeration may also be used to target these counties.
Item non-response will be handled in one of four ways. First, deterministic imputation will be employed whenever the missing value can be derived from other cells on the questionnaire. Second, previously reported data from either a recent NASS survey or the previous census, will be imputed, when appropriate. Third, a nearest neighbor donor (farm of similar type, size, and location) will be found and a value or relationship from the donor will be used for the recipient. When all of these automated options fail, the problem will be referred to a statistician for resolution. The statistician will utilize knowledge from training, existing agricultural knowledge, computer program analysis tools, and on some occasions will conduct callbacks to the respondent. Unit non-response is handled through weighting, as described in item 2, above.
Relative standard errors for U.S. fully adjusted estimates of number of farms for major demographic items in 2012 ranged from 0.7 to 22.5 (see Table B in Appendix A of the 2012 Census of Agriculture United States Summary and State Data, Volume 1, Part 51). In 2017, NASS will survey the same number of segments for Census adjustments. Thus, the relative standard errors are expected to be similar to 2012.
4. Describe any tests of procedures or methods to be undertaken.
NASS used 2012 Census data to identify problem areas of the questionnaire by looking at edit and imputation rates by question and section. Comments and suggestions from field staff were also reviewed for potential problems with data collection and questionnaires.
NASS conducted a total of 100 cognitive interviews for the Census of Agriculture questionnaires, focusing on new content as well as alternate ways to collect commodity data. In addition, NASS conducted a content test in 2015 (OMB No. 0535-0243) with a nationwide sample of 30,000 operations, using six versions of the questionnaire. Alternate versions of the questionnaire were tested to look at the impact of section order, differences in collecting commodity data, and respondent comprehension of new content. Content test results were used to determine the final questionnaire content and design.
2017 processing systems have been updated based on experiences from the 2012 Census. All processing systems will be tested using previous Census data, as well as data from the 2015 content test.
5. Provide the name and telephone number of individuals consulted on statistical aspects of the design and the name agency unit, contractor(s), grantee(s), or others who will actually collect and/or analyze the information for the agency.
Several NASS units contribute to developing census methodology, each containing staff members with prior Census experience. Contributing senior staff and unit leaders are:
Mark Apodaca, Chief, Sampling, Editing, and Imputation Methodology Branch: (202)720-4008
Jeff Bailey, Chief, Summary, Estimation, and Disclosure Methodology Branch: (202)720-4008
Dan Beckler, Chief, Standards and Survey Methodology Branch: (202)720-4008
Donald Buysse, Chief, Census Planning Branch: (202)690-8747
Troy Joshua, Chief, Environmental, Economics, and Demographics Branch: (202)720-6146
Jaki McCarthy, Senior Survey Methodologist: (202)690-2389
Linda Young, Director, Research and Development Division: (202)690-1401
June 2016
Revised October 2016
| File Type | application/vnd.openxmlformats-officedocument.wordprocessingml.document |
| Author | hancda |
| File Modified | 0000-00-00 |
| File Created | 2021-01-23 |